Messages in this thread | | | From | "Ulrich Windl" <> | Date | Fri, 8 Aug 1997 09:58:31 +0200 | Subject | scsi.c, aic7xxx: problems in 2.0.28 |
| |
I experienced some severe problems wen trying to read the defect list of my SCSI harddisk wirh "scsiinfo -d". The harddisk is a Conner 1080S with a rather large defect list supplied by the manufacturer.
scsiinfo -d causes a "aborting command due to timeout... Read Defect Data" From that moment on the system seems to execute never-ending SCSI bus resets. I have a AHA 2940 with BIOS 1.21 that otherwise works well.
I have enabled "DEBUG_TIMEOUT" in scsi.c and booted Linux with "single debug". Interestingly in this constellation the scsiinfo at least terminates but with varying results (e.g. defect list is said to be empty).
I have written down some of the possibly "interesting" messages:
aic7xxx: (abort) Aborting scb 2, TCL=0/0/0 ... SCSI host 0 abort (pid 9363) timed out - resetting aic7xxx: (reset) target/channel 0/0 (abort_reset) scb state 0x1, Data-in pahse, SCSISIGI=0xc6 (abort_scb) asserted ATN - bus device reset in message buffer SCSI host 0 channel 0 reset (pid 9363) timed out - trying harder aic7xxx: (reset) target/channel 0/0 (reset_device) target/channel -1/A, active_scb 2 (match_scb) comparing target/channel -1/A to scb 0/A (reset_channel) Resetting current channel A (reset_channel) Channel reset, sequencer restarted (done_aborted_scbs) Aborting scb 2, TCL=0/0/0
I noticed that the value for wait_for_request alsways stayed the same, namedly at 001dff00. Also I noticed that during bus or device reset the system had seriously delayed console switching (several seconds). Is that a bug?
What is really interesting is that even that I was able to get scsiinfo finished, the system was unable to read any further block from the SCSI disk. I think that is the real bug. Unfortunately I can't tell whether the problem is in the aic7xxx driver or in scsi.c.
After such a disk read error I also got spurious messages from the ext2 filesystem.
For aic7xxx: the comparison of -1/A and 0/A seems odd to me. also sometimes channels seem to be printed as a digit and sometimes as a letter.
Hopefully this report is useful to someone. I'm not subscribed to thes lists, and I'm on holiday for another week...
Ulrich Windl
| |